携手火山引擎，顺丰科技用AI重塑供应链

Core Insights - The application of large models in long-chain and complex collaborative supply chains is one of the most effective areas for efficiency improvement [1] - SF Technology is actively exploring best practices in the era of large models, collaborating with Volcano Engine to support the digitalization of logistics [1][2] - In 2024, SF Technology will launch the "Fengyu" vertical large model for the logistics industry, utilizing Volcano Engine's AI training and inference services across over 20 business scenarios [1][2] Group 1: Model Development and Application - The "Fengyu" model family includes language, voice, and multimodal models, applied in various logistics functions such as marketing, customer service, and international customs [2] - The training process of the "Fengyu" model is accelerated through Volcano Engine's machine learning platform, achieving full-link efficiency from training to inference [3][4] Group 2: Performance and Cost Efficiency - The veTuner training framework enhances the performance of the "Fengyu" model by over 30% compared to open-source frameworks, with support for various reinforcement learning algorithms [4] - The xLLM inference framework increases the throughput capacity of the "Fengyu" model by up to 5 times, effectively managing high concurrency during peak business periods [4] - SF Technology is addressing rising inference costs by utilizing Volcano Engine's AI cloud-native inference suite, achieving significant efficiency improvements in model deployment and application [4]